Quick Stata Guide

Quick Stata Guide
by Liz Foster
Table of Contents
Part 1: Top Ten Stata Commands
describe
generate
regress
scatter
sort
summarize
table
tabulate
test
ttest
1
1
1
3
4
5
5
6
8
10
11
Part 2: Prefixes and Notes

by var:
capture
use of the *
explanation of data set
14
14
14
15
16
Part 3: More Examples

interaction terms
regression line
17
17
18
Appendix: Key Terms and Concepts
20
Quick Stata Guide
Top Ten Stata Commands
Part 1: Top Ten Stata Commands

describe
This command tells you information about the variables in your dataset how big they are,
what they represent, units, what different codes stand for if this information is available.
Example
. describe
Contains data from example.dta
obs:
281
Child Support Awards Santa

Clara County California
18 Nov 2004 15:52
vars:
5
size:
4,496 (99.6% of memory free)
------------------------------------------------------------------------------storage display
value
variable name
type
format
label
variable label
------------------------------------------------------------------------------award
int
%8.0g
Child support award
earndad
float %9.0g
Father's monthly earnings
earnmom
float %9.0g
Mother's monthly earnings
nkids
byte
%8.0g
Number of kids
petmom
byte
%8.0g
yesno
Was it the mother who
petitioned for divorce?
------------------------------------------------------------------------------Sorted by:
Options
You can select only certain variables by listing them, for example:
describe earnmom earndad
generate
This command generates new variables. In particular, it can generate dummy variables and
interaction terms. It can be abbreviated gen.
Example
. gen richmom = (earnmom >= 2500)
. table richmom, c(freq min earnmom max earnmom mean earnmom)
---------------------------------------------------------------------richmom |
Freq.
min(earnmom)
max(earnmom) mean(earnmom)
----------+----------------------------------------------------------0 |
239
0
2491.67
1205.992
1 |
42
2500
5250
2950.984
----------------------------------------------------------------------
This creates a new binary variable equal to 1 if the mother earns more than $2500 a month.
Or, we could generate a variable that indicated whether the mother earned more than the
father:
gen richermom = (earnmom > earndad)
If you want to see whether child support is a quadratic function of the number of children,
Quick Stata Guide
rather than linear, you need to add an nkids2 term.

. gen nkidssq = nkids * nkids
. reg award nkids nkidssq, r
Regression with robust standard errors
Number of obs
F( 2,
278)
Prob > F
R-squared
Root MSE
=
=
=
=
=
281
31.92
0.0000
0.1921
218.1
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------nkids |
446.319
111.7826
3.99
0.000
226.2711
666.3669
nkidssq | -80.59451
32.09875
-2.51
0.013
-143.782
-17.40702
_cons | -106.9748
85.40495
-1.25
0.211
-275.0973
61.14779
------------------------------------------------------------------------------
The coefficient on the squared term is significant, so the quadratic form fits the data better.
To see whether the effect of the mother being the petitioner is different for mothers who earn
more than their husbands, we need an interaction term richermom * petmom.
. gen richermom_X_petmom = richermom * petmom
. reg award richermom petmom richermom_X_petmom, r
Number of obs
F( 3,
277)
Prob > F
R-squared
Root MSE
=
=
=
=
=
281
17.06
0.0000
0.1970
217.83
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------richermom | -257.8725
66.34101
-3.89
0.000
-388.4691
-127.2759
petmom | -150.4746
64.45085
-2.33
0.020
-277.3503
-23.5989
richermom_~m |
87.16457
74.13835
1.18
0.241
-58.78159
233.1107
_cons |
596.4483
57.13669
10.44
0.000
483.971
708.9256
------------------------------------------------------------------------------
The interaction term is not significant.

Options
The command gen can be combined with the command tab to generate a set of indicator
variables for the categories of a category variable. For example:
. tab nkids, gen(nkids_)
Number of |
kids |
Freq.
Percent
Cum.
------------+----------------------------------1 |
143
50.89
50.89
2 |
117
41.64
92.53
3 |
20
7.12
99.64
4 |
1
0.36
100.00
------------+-----------------------------------
Quick Stata Guide

Total |

281
100.00
. sum nkids_*
Variable |
Obs
Mean
Std. Dev.
Min
Max
-------------+-------------------------------------------------------nkids_1 |
281
.5088968
.5008128
0
1
nkids_2 |
281
.4163701
.4938359
0
1
nkids_3 |
281
.0711744
.2575746
0
1
nkids_4 |
281
.0035587
.059655
0
1
This creates four new indicator variables. For example, nkids_2 is equal to 2 if the family has
two children, and 0 otherwise. We can now regress the child support award on the number of
children in the most flexible way possible without assuming the relationship to be linear or
quadratic.
. reg award nkids_*, r
Number of obs =
281
F( 3,
277) =
54.72
Prob > F
= 0.0000
R-squared
= 0.1954
Root MSE
= 218.04
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------nkids_1 |
60.06993
13.61372
4.41
0.000
33.27045
86.86941
nkids_2 |
258.4444
23.15878
11.16
0.000
212.8549
304.034
nkids_3 |
334.95
74.62313
4.49
0.000
188.0495
481.8505
nkids_4 | (dropped)
_cons |
200
.
.
.
.
.
------------------------------------------------------------------------------
For help in interpreting these results, see the page for test.
regress
This command runs an OLS regression. The first variable is the dependant one (Y) the
following are the independent ones (Xs). Can be abbreviated reg.
Example
. reg award nkids
Source |
SS
df
MS
-------------+-----------------------------Model | 2730761.85
1 2730761.85
Residual | 13636695.1
279 48877.0432
-------------+-----------------------------Total | 16367456.9
280 58455.2033
Number of obs
F( 1,
279)
Prob > F
R-squared
Adj R-squared
Root MSE
=
=
=
=
=
=
281
55.87
0.0000
0.1668
0.1639
221.08
-----------------------------------------------------------------------------award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------nkids |
154.1658
20.62522
7.47
0.000
113.565
194.7666
_cons |
120.0708
34.95281
3.44
0.001
51.26609
188.8755
------------------------------------------------------------------------------
Quick Stata Guide
Options
The option , r is added so that Stata allows for heteroskedasticity and calculates the correct
standard errors. According to Watson, you should always use it. It changes the format of the
output a little:
. reg award nkids, r
Number of obs
F( 1,
279)
Prob > F
R-squared
Root MSE
=
=
=
=
=
281
35.12
0.0000
0.1668
221.08
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------nkids |
154.1658
26.01487
5.93
0.000
102.9554
205.3761
_cons |
120.0708
36.85967
3.26
0.001
47.51243
192.6292
------------------------------------------------------------------------------
So we have the result that award = 120.1 + 154.2 * nkids with standard errors of 26.0 and
36.9 on the two coefficients.
Stata no longer automatically displays the adjusted R-squared. To make Stata display it, use:
display _result(8)
scatter
Produces basic scatter plots of data.
Example
. scatter earnmom award
Quick Stata Guide
Options
Add more variables. The last variable will always be on the x-axis, the other variables on the
y-axis, represented by different colored dots.
. scatter earnmom earndad award
There are dozens of other options read Stata help. If you want to change something about
the scatter plot, you can.
sort
This commands sorts your data by the values of a specific variable. It must be run before you
can use the prefix by :
Example
The command
sort nkids
produces no output, but if you now run describe it will tell you that your dataset is sorted
by nkids.
summarize
If run with no arguments, this command produces a basic summary of every variable in your
data set. It may be abbreviated sum.
Quick Stata Guide
Example
. summarize
Variable |
Obs
Mean
Std. Dev.
Min
Max
-------------+-------------------------------------------------------award |
281
362.0178
241.7751
0
1600
earndad |
281
1363.912
2514.409
0
28333.33
earnmom |
281
1466.809
962.5245
0
5250
nkids |
281
1.569395
.6405823
1
4
petmom |
281
.7793594
.4154184
0
1
This data set has 5 variables, called award, earndad, earnmom, nkids, and petmom. The first
column Obs tells you the number of observations you have for each variable here we have
281 for each. The second column Mean tells you the average value of each variable in the
dataset. The third column Std. Dev. tells you the standard deviation of the variable. The fourth
and fifth columns Min and Max tell you the smallest and largest value of the variable in the
dataset.
Options
You can add a list of variables to produce summary stats for those variables only. For
example:
. summarize earnmom earndad
Variable |
Obs
Mean
Std. Dev.
Min
Max
-------------+-------------------------------------------------------earnmom |
281
1466.809
962.5245
0
5250
earndad |
281
1363.912
2514.409
0
28333.33
You can add the option , detail to produce more detailed statistics for one or more
variables.
. summarize earnmom, detail
earnmom
------------------------------------------------------------Percentiles
Smallest
1%
0
0
5%
0
0
10%
0
0
Obs
281
25%
900
0
Sum of Wgt.
281
50%
75%
90%
95%
99%
1500
2100
2666.67
2950
4083.33
Largest
3700
4083.33
4158.33
5250
Mean
Std. Dev.
1466.809
962.5245
Variance
Skewness
Kurtosis
926453.4
.2368765
3.094632
table
When given a list of variables, produces tables showing the frequency of combinations of
values of those variables.
Quick Stata Guide
Examples
One variable:
. table nkids
---------------------nkids |
Freq.
----------+----------1 |
143
2 |
117
3 |
20
4 |
1
----------------------
Two variables:
. table nkids petmom
---------------------|
petmom
nkids |
0
1
----------+----------1 |
26
117
2 |
32
85
3 |
4
16
4 |
1
----------------------
Three variables:
. table award petmom nkids
---------------------------------------------------------------|
nkids and petmom
| ---- 1 ------ 2 ------ 3 ------ 4 --award |
0
1
0
1
0
1
0
1
----------+----------------------------------------------------0 |
1
10
3
50 |
1
1
60 |
1
63 |
1
75 |
1
85 |
1
90 |
1
100 |
2
2
101 |
1
...
Options
The power of table lies in its ability to present a wide range of other statistics in these tables
instead of simple frequencies. This is done by an option of the form
, c(stat1 var1 stat2 var2 ... )
. table nkids petmom, c(mean earnmom)
-----------------------------|
petmom
nkids |
0
1
----------+------------------1 | 1811.013 1610.766
2 | 1099.323
1324.06
Quick Stata Guide
3 | 605.8325 1523.801
4 |
2100
------------------------------
This table presents the average mother's earnings broken down by number of kids and
whether the mother was the petitioner. The next table gives the average value and standard
deviation of the award.
. table nkids
petmom, c(mean award sd award)
-----------------------------|
petmom
nkids |
0
1
----------+------------------1 | 308.846
249.231
| 182.9812 156.0083
|
2 | 578.906
413.094
| 302.8179 211.4426
|
3 |
478.75
549
| 274.099 360.9759
|
4 |
200
|
------------------------------
Some other statistics you may find useful include freq (frequency), sum, median, max and
min. Note that freq is not followed by a variable name.
Note
table and tabulate overlap greatly in what they do. In particular, the following two
commands
tabulate var1, sum(var2)
table var1, c(mean var2 sd var2 freq)
produce exactly the same information. table allows you much more flexibility in exactly
what information you present and the number of variables you can work with at once.
tabulate is quicker significantly quicker for largish datasets.
tabulate
When this command is give one variable, it creates a table showing the values that variable
takes on. It can be abbreviated tab.
Example
. tabulate nkids
nkids |
Freq.
Percent
Cum.
------------+----------------------------------1 |
143
50.89
50.89
2 |
117
41.64
92.53
3 |
20
7.12
99.64
4 |
1
0.36
100.00
------------+----------------------------------Total |
281
100.00
Quick Stata Guide
This shows that nkids takes on values from 1 to 4. Freq (Percent) tells you the number
(percentage) of observations with each number of kids. Cum. tells you the total percentage of
observations with less than or equal to that number of kids.
Options
The option , sum(var) can be added to the end, so that instead of just giving the
percentage / cumulative percentages for each value, Stata gives you the average value of
another variable.
. tab nkids, sum(earnmom)
|
Summary of earnmom
nkids |
Mean
Std. Dev.
Freq.
------------+-----------------------------------1 |
1647.1741
950.28916
143
2 |
1262.5931
940.80927
117
3 |
1340.2075
979.98086
20
4 |
2100
0
1
------------+-----------------------------------Total |
1466.8094
962.52452
281
Thus we see that for families with 2 kids, the average mother's earnings are $1262.59.
Two variables can be given to create a table that shows how different variables are distributed
together.
. tab nkids petmom
|
petmom
nkids |
0
1 |
Total
-----------+----------------------+---------1 |
26
117 |
143
2 |
32
85 |
117
3 |
4
16 |
20
4 |
0
1 |
1
-----------+----------------------+---------Total |
62
219 |
281
The numbers in the table are frequencies. There were 85 families with two kids where the
mother was the petitioner.
These two options can be used together.
. tab nkids petmom, sum(award)
Means, Standard Deviations and Frequencies of award
|
petmom
nkids |
0
1 |
Total
-----------+----------------------+---------1 | 308.84615 249.23077 | 260.06993
| 182.98124 156.00825 | 162.20165
|
26
117 |
143
-----------+----------------------+---------2 | 578.90625 413.09412 | 458.44444
| 302.81788 211.44264 | 249.78092
|
32
85 |
117
-----------+----------------------+---------3 |
478.75
549 |
534.95
Quick Stata Guide
| 274.09898
360.9759 | 339.94868
|
4
16 |
20
-----------+----------------------+---------4 |
.
200 |
200
|
.
0 |
0
|
0
1 |
1
-----------+----------------------+---------Total | 459.19355 334.50685 | 362.01779
| 284.94871
221.1655 | 241.77511
|
62
219 |
281
test
After a regression, this command can be used to test various hypotheses about the
coefficients of the regression, including an F-test for joint significance.
Example
. capture tab nkids, gen(nkids_)
Number of obs
F( 3,
277)
Prob > F
R-squared
Root MSE
=
=
=
=
=
281
54.72
0.0000
0.1954
218.04
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------nkids_1 |
60.06993
13.61372
4.41
0.000
33.27045
86.86941
nkids_2 |
258.4444
23.15878
11.16
0.000
212.8549
304.034
nkids_3 |
334.95
74.62313
4.49
0.000
188.0495
481.8505
nkids_4 | (dropped)
_cons |
200
.
.
.
.
.
-----------------------------------------------------------------------------. test nkids_1 nkids_2 nkids_3
( 1)
( 2)
( 3)
nkids_1 = 0
nkids_2 = 0
nkids_3 = 0
F(
3,
277) =
Prob > F =
54.72
0.0000
This tells you that the indicator variables of the number of kids are jointly significant not
surprising, since they're all individually highly significant.
Options
In order to use * to avoid having to type out all the names of the indicator random
variables, you can use the command testparam.
. testparam nkids_*
( 1)
( 2)
nkids_1 = 0
nkids_2 = 0
10
Quick Stata Guide

( 3)
( 4)
nkids_3 = 0
nkids_4 = 0
Constraint 4 dropped
F(
3,
277) =
Prob > F =
54.72
0.0000
You can test basically any statement about the coefficients. For example the child support
award is increasing the father's earnings and decreasing in mother's earnings, but are the two
effects of the same size?
. reg award earnmom earndad, r
Number of obs
F( 2,
278)
Prob > F
R-squared
Root MSE
=
=
=
=
=
281
5.06
0.0069
0.2256
213.53
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------earnmom | -.0392945
.0144494
-2.72
0.007
-.0677385
-.0108504
earndad |
.0437977
.016691
2.62
0.009
.0109409
.0766545
_cons |
359.919
25.24921
14.25
0.000
310.2151
409.623
-----------------------------------------------------------------------------. test earnmom = -earndad
( 1)
earnmom + earndad = 0
F(
1,
278) =
Prob > F =
0.07
0.7914
Answer: could well be.
ttest
Performs a statistical test as to whether a variable has a specified mean, whether two
variables are equal, or whether the mean of one variable is equal across values of another
variable.
Examples
First case: test whether in half the cases the mother is the petitioner i.e., whether the mean
of petmom is 0.5.
. ttest petmom==0.5
One-sample t test
-----------------------------------------------------------------------------Variable |
Obs
Mean
Std. Err.
Std. Dev.
---------+-------------------------------------------------------------------petmom |
281
.7793594
.0247818
.4154184
.7305772
.8281417
-----------------------------------------------------------------------------Degrees of freedom: 280
11
Quick Stata Guide

Ho: mean(petmom) = 0.5
Ha: mean < 0.5

t = 11.2728
P < t =
1.0000
Ha: mean != 0.5

t = 11.2728
P > |t| =
0.0000
Ha: mean > 0.5

t = 11.2728
P > t =
0.0000
This presents data on the values of petmom the mean, standard deviation and a 95%
confidence interval for the mean. It also specifically tests whether the mean of petmom is
equal to 0.5 and finds that against the alternative that it's not equal to 0.5 (middle column on
the bottom) the t-stat is 11.27 which corresponds to a p-value of 0 so we reject the
hypothesis that in half the cases the petitioner is the mother.
Now we test whether for each family, mother's earnings are equal to father's earnings on
average. This basically creates a variable that for each family is the difference in earnings, and
tests whether it has mean 0.
. ttest earnmom == earndad
Paired t test
-----------------------------------------------------------------------------Variable |
Obs
Mean
Std. Err.
Std. Dev.
---------+-------------------------------------------------------------------earnmom |
281
1466.809
57.4194
962.5245
1353.781
1579.838
earndad |
281
1363.912
149.997
2514.409
1068.647
1659.177
---------+-------------------------------------------------------------------diff |
281
102.8973
158.193
2651.798
-208.5013
414.2959
-----------------------------------------------------------------------------Ho: mean(earnmom - earndad) = mean(diff) = 0
Ha: mean(diff) < 0
t =
0.6505
P < t =
0.7420
Ha: mean(diff) != 0
t =
0.6505
P > |t| =
0.5159
Ha: mean(diff) > 0

t =
0.6505
P > t =
0.2580
Here, we can't reject the hypothesis that mothers and father have the same earnings on
average.
Now we test whether or not awards are equal in cases where the mother is the petitioner
versus cases where she isn't. We tell Stata to break the data up into groups based on the value
of petmom and test if the mean of award is the same in all these groups.
. ttest award, by(petmom)
Two-sample t test with equal
variances
-----------------------------------------------------------------------------Group |
Obs
Mean
Std. Err.
Std. Dev.
---------+-------------------------------------------------------------------0 |
62
459.1935
36.18852
284.9487
386.8301
531.557
1 |
219
334.5068
14.94498
221.1655
305.0517
363.962
---------+-------------------------------------------------------------------combined |
281
362.0178
14.42309
241.7751
333.6263
390.4093
---------+-------------------------------------------------------------------diff |
124.6867
34.03465
57.68939
191.684
-----------------------------------------------------------------------------Degrees of freedom: 279
Ho: mean(0) - mean(1) = diff = 0
12
Quick Stata Guide

Ha: diff < 0
t =
3.6635
P < t =
0.9999

Ha: diff != 0
t =
3.6635
P > |t| =
0.0003
Ha: diff > 0

t =
3.6635
P > t =
0.0001
Options
Note that in the last example, Stata assumes equal variances (homoskedasticity). If we want
to allow that the variance of the award might differ depending on whether or not the mother
is the petitioner we want to add the option unequal as in:
ttest award, by(petmom) unequal
This accomplishes the same thing as adding the option , r to a regression command.
The prefix by var: can also be used to break up the data further see the explanation for
this prefix.
13
Quick Stata Guide
Prefixes and Notes
Part 2: Prefixes and Notes

by :
If you have data sorted by the values of a variable, you can run a command separately for
each value of the variable. Simply add by var: before the command.
Example
. sort petmom
. by petmom: reg award nkids, r
_______________________________________________________________________________
-> petmom = no
Number of obs =
F( 1,
60) =
Prob > F
=
R-squared
=
Root MSE
=
62
10.17
0.0023
0.1446
265.73
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------nkids |
179.6584
56.34751
3.19
0.002
66.94663
292.3702
_cons |
163.6265
84.32099
1.94
0.057
-5.040636
332.2935
-----------------------------------------------------------------------------_______________________________________________________________________________
-> petmom = yes
Number of obs
F( 1,
217)
Prob > F
R-squared
Root MSE
=
219
=
23.93
= 0.0000
= 0.1755
= 201.28
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------nkids |
142.4462
29.12188
4.89
0.000
85.0483
199.8442
_cons |
114.0079
40.80712
2.79
0.006
33.57882
194.4369
------------------------------------------------------------------------------
This shows us that when the father is the petitioner for divorce, each child increases the
award by about $179.66 while if the mother is the petitioner for divorce, each child
increases the award by about $142.44. (However, if you look at the confidence intervals,
they overlap significantly.)
Note
You must have sorted the data by the variable you wish to use before using by :, if not,
you'll get the error "not sorted".
capture
Makes Stata suppress the normal output from a command.
14
Quick Stata Guide
Prefixes and Notes
Example
The command
capture tab earnmom, gen(em_cat)
will create a indicator variable for each category of mother's earnings we have in the data
set, but without printing out each level of earnings.
Note
The command quietly seems to do exactly the same thing.
*
The asterisk can be used to include a set of variables without having to type them all out.
In many commands (sum, describe and reg) instead of a variable use var* to include
all variables whose names start with var.
Example
. tab nkids, gen(nkids_)
Number of |
kids |
Freq.
Percent
Cum.
------------+----------------------------------1 |
143
50.89
50.89
2 |
117
41.64
92.53
3 |
20
7.12
99.64
4 |
1
0.36
100.00
------------+----------------------------------Total |
281
100.00
Number of obs
F( 3,
277)
Prob > F
R-squared
Root MSE
=
=
=
=
=
281
54.72
0.0000
0.1954
218.04
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------nkids_1 |
60.06993
13.61372
4.41
0.000
33.27045
86.86941
nkids_2 |
258.4444
23.15878
11.16
0.000
212.8549
304.034
nkids_3 |
334.95
74.62313
4.49
0.000
188.0495
481.8505
nkids_4 | (dropped)
_cons |
200
.
.
.
.
.
------------------------------------------------------------------------------
This is particularly useful when you have a lot of indicators variables.

Notes
Be sure to type nkids_* and not nkids* so that Stata does not include the original
variable nkids
This does not work with the command test, but you can use the command testparam
15
Quick Stata Guide
Prefixes and Notes
that does an F-test for joint significance and takes a variable expression with *.
Explanation of the Dataset

The dataset used in the examples is based on a data set we used in 508c last year. It
presents information on 281 child support cases in Santa Clara county, California. In all
these cases, the mother has physical custody of the children, and the child support
payment is to be paid by the father. It has 5 variables:
award the amount of the child support that was awarded
earnmom the mother's monthly earnings
earndad the father's monthly earnings
nkids the number of children in the family
petmom a binary variable equal to 1 if the mother petitioned for divorce and 0 if the
father petitioned for divorce.
16
Quick Stata Guide
More Examples
Part 3: More Examples

Interaction Term
First we'll generate a dummy variable to indicate if the mother earns more than the
father. Then we'll look at the average award for each group of people.
. gen richermom = (earnmom > earndad)
. table richermom petmom, c(mean award)
---------------------------|Was it the mother
| who petitioned
|
for divorce?
richermom |
no
yes
----------+----------------0 | 596.448 445.974
1 | 338.576 275.266
----------------------------
What happens if we regress the child support award on whether or not the mother was
the petitioner and whether or not she earns more than the father?
. reg award richermom petmom, r
Number of obs
F( 2,
278)
Prob > F
R-squared
Root MSE
=
=
=
=
=
281
25.18
0.0000
0.1915
218.18
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------richermom | -191.3874
29.79194
-6.42
0.000
-250.0339
-132.741
petmom | -101.5843
35.1742
-2.89
0.004
-170.8259
-32.34273
_cons |
561.061
39.18008
14.32
0.000
483.9337
638.1884
------------------------------------------------------------------------------
Based on these regression coefficients, we can predict how much should be awarded to
each type of mother.
Mother petitioned?
yes
no
Mother earns
yes
561.061
561.061 - 101.5843 =
more?
459.4767
no
561.061 - 191.3874 =
561.061 - 101.5843 - 191.3874 =
369.6736
268.0893
Note that these numbers are close the averages we calculated, but not exact. In
particular, from the first table, it looks like it makes more of a difference who petitions
for divorce when the father is richer. To allow for this type of phenomenon, we need an
interaction term.
. gen richer_pet = richermom * petmom
. reg award richermom petmom richer_pet, r
17
Quick Stata Guide
More Examples
Number of obs
F( 3,
277)
Prob > F
R-squared
Root MSE
=
=
=
=
=
281
17.06
0.0000
0.1970
217.83
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------richermom | -257.8725
66.34101
-3.89
0.000
-388.4691
-127.2759
petmom | -150.4746
64.45085
-2.33
0.020
-277.3503
-23.5989
richer_pet |
87.16457
74.13835
1.18
0.241
-58.78159
233.1107
_cons |
596.4483
57.13669
10.44
0.000
483.971
708.9256
------------------------------------------------------------------------------
Now if we recalculate our table of predicted values from the regression result, we exactly
replicate the sample averages.
Mother petitioned?
yes
Mother
earns
more?
yes
596.4483
no
596.4483 - 257.8725 =
338.5758
no
596.4483 - 150.4746 =
445.9737
596.4483 - 257.8725 - 150.4746 + 87.16457 =
275.26577
Notice however that the interaction term is not statistically significant, so that difference
we saw may just be an artifact of the data.
Graphing the Regression Line

If we look at the data, we see that there are two outliers of fathers who make more than
$15,000 a month that's more than $180,000 a year.
We're going to drop these two observations to make our graphs easier to see. Note that
this does change the regression results, so if you were really doing this you would want to
think before just dropping them.
. drop if earndad > 10000
. reg award earndad, r
Number of obs
F( 1,
277)
Prob > F
R-squared
Root MSE
=
=
=
=
=
279
43.07
0.0000
0.2979
200.91
-----------------------------------------------------------------------------|
Robust
award |
Coef.
Std. Err.
t
P>|t|
-------------+---------------------------------------------------------------earndad |
.0770272
.0117365
6.56
0.000
.0539231
.1001313
_cons |
265.7285
14.62653
18.17
0.000
236.9352
294.5218
-----------------------------------------------------------------------------. predict award_hat
18
Quick Stata Guide
More Examples
(option xb assumed; fitted values)

. scatter award earndad || scatter award_hat earndad, c(l) sort m(i)
(note that's the letter 'l' not the number '1' after the c.)
And we get this result:
19
Quick Stata Guide
Appendix
Key Terms and Concepts

Note: Page numbers refer to Stock and Watson.
A regression coefficient is significant if
its value is more than twice its standard error (quick rule for significance at 5%
level)
its t-stat is greater than 2.58 / 1.96 / 1.64 (significance at 1% / 5% / 10% level)
its p-value is less than 0.01 / 0.05 / 0.10 (significance at 1% / 5% / 10% level)
(p 112-114)
A 95% confidence interval for a coefficient is the estimated value 1.96
standard error (p 117-118)
Interpretation of a regression coefficient:
linear-linear (Y = 0 + 1X): an increase of one unit of X is associated with an
increase of 1 units of Y
log-linear (ln Y = 0 + 1X): an increase of one unit of X is associated with a
100 1 % increase in Y
linear-log (Y = 0 + 1 ln X): a 1% increase in X is associated with a 0.01 1
unit increase in Y
log-log (ln Y = 0 + 1 ln X): a 1 % increase in X is associated with a 1 %
increase in Y (p 210-214)
In a regression Y = 0 + 1X: Y is the dependent variable and X is the
independent variable. (p 94)
The R2 of a regression is the fraction of variation in the dependent variable (Y)
that is explained by the regression. (The adjusted- R2 or R-bar-squared is the
same thing with a small technical adjustment which means you can compared it across
regressions with different numbers of variables.) (p 122, 176)
The standard error of the regression (SER) is an estimator of the standard error
of the regression error u. It has the same units as Y. ESS, TSS and SSR are less
important. (p 122-123)
An F-test tests whether multiple coefficients could all be 0.
If the p-value is less than 0.05 (or the F-stat is greater than the critical value)
then we can reject the possibility that all the coefficients are 0.
If the p-value is greater than 0.05 then we cannot reject that all the
coefficients might be 0. (p 165-169, inside back cover)
There is evidence of a non-linear effect of X on Y if when the variable X2 is added to
the regression, its coefficient is significant. (p 206)
For a regression Y = 0 + 1X + 2W + 3XW : there is evidence for an interaction of
between X and W (or that the effect of X on Y depends on W) if the coefficient of
the interaction term X W is significant. (If X is included as X, X2, X3
interaction terms between W and all the powers of X must be included and an F-test
done on all the interaction terms) (p 218-229)
20
Quick Stata Guide
Appendix
Difference in Differences
The following table gives average or predicted values of Y for 4 groups:
X=1
X=0
W=1
W=0
The difference in differences is (1 - 3) - (2 - 4). This should be the same as the value
of 3 in the regression Y = 0 + 1X + 2W + 3XW
The error term is
homoskedastic if the variance of Y is the same for different values of X (ie, the
variance of test scores is the same for kids in small classes as large classes).
heteroskedastic is the variance of Y is different for different values of X.
If you assume homoskedasticity wrongly, your standard errors will be too small (but
your coefficient unbiased). If you allow for heteroskedasticity, your standard error will
be right even if the error term is really homoskedastic. (p 124-126, 129)
You have omitted variable bias is there is some factor that is correlated with your
independent variables (X etc) and influences the dependent variable Y but is not
included in the regression.
This means that your coefficients are, on average, wrong.
If X is truly randomly assigned then you don't have OVB. (p 144-147)
To derive an OLS estimator for a regression Y = f(X,) + u where you get to pick :
1. Set up what you want to minimize : (Y f(X, ) )2
2. Take the derivative with respect to and set = 0.
3. Solve for .
To prove an estimator -hat is unbiased
1. Use Y = f(X,) + u to write -hat in terms of , X and u preferably as + some
expression in X and u.
2. Take expectations.
3. Use LIE to replace u with E[u | X].
4. Use the first assumption to say that E[u | X] = 0 and simplify. (p 135-137)
Remember:
E[a+bX] = a + bE[X]
var(a+bX) = b2 var(X)) var(X) = E[(X-)2] = E[X2] (E[X])2
If A and B are normally distributed with means ma and mb and variance sa and sb
(standard errors ea and eb) then A-B is normally distributed with mean ma-mb and
variance sa+sb (standard error (ea + eb) )
21

Quick Stata Guide

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Quick Stata Guide

Uploaded by

Copyright:

Available Formats

Quick Stata Guide

Part 2: Prefixes and Notes

Part 3: More Examples

Appendix: Key Terms and Concepts